Machine Learning: Course Overview CS 760@UW-Madison Class - - PowerPoint PPT Presentation

Machine Learning: Course Overview

CS 760@UW-Madison

Class enrollment

• typically the class was limited to 30
• we’ve allowed ~100 to register
• the waiting list full
• unfortunately, many on the waiting

list will not be able to enroll

• but CS760 will be offered in the

next semester!

Instructor

• Yingyu Liang

email: yliang@cs.wisc.edu

• ffice hours: Mon 4-5pm
• ffice: 6393 Computer Sciences

TA

• Ying Fan

email: yingfan@cs.wisc.edu

• ffice hours: Tues 1-2pm, Thu 1-2pm
• ffice: CS 1351

Monday, Wednesday and Friday?

• we’ll have ~30 lectures in all, just like a standard TR class
• will push the lectures forward (finish early, leave time for

projects and review)

• see the schedule on the course website:

http://pages.cs.wisc.edu/~yliang/cs760_spring20

Course emphases

• a variety of learning settings: supervised learning,

unsupervised learning, reinforcement learning, active learning, etc.

• a broad toolbox of machine-learning methods: decision trees,

nearest neighbor, neural nets, Bayesian networks, SVMs, etc.

• some underlying theory: bias-variance tradeoff, PAC learning,

mistake-bound theory, etc.

• experimental methodology for evaluating learning systems:

cross validation, ROC and PR curves, hypothesis testing, etc.

Two major goals

• 1. Understand what a learning system should do
• 2. Understand how (and how well) existing systems work

Course requirements

• 7-8 homework assignments: 30%
• programming
• computational experiments (e.g. measure the effect of

varying parameter x in algorithm y)

• some written exercises
• Midterm Exam #1: 20%
• Midterm Exam #2: 20%
• final project: 30%
• project group: 3-5 people

Expected background

• CS 540 (Intro to Artificial Intelligence) or equivalent
• good programming skills
• probability
• linear algebra
• calculus, including partial derivatives

Programming languages

• for the programming assignments, you can use

C C++ Java Perl Python R Matlab

• suggest: Python
• programs must be callable from the command line and must

run on the CS lab machines (this is where they will be tested during grading!)

Course readings

Recommend to get one of the following books

• Pattern Recognition and Machine Learning. C. Bishop. Springer, 2011.
• Machine Learning: A Probabilistic Perspective. K. Murphy. MIT Press,

2012.

• Understanding Machine Learning: From Theory to Algorithms. S.

Shalev-Shwartz, S. Ben-David. Cambridge University press, 2014.

Course readings

• the books can be found online or at Wendt Commons

Library

• additional readings will come from online articles, surveys,

and chapters

• will be posted on course website

Machine Learning Examples

What is machine learning?

• “A computer program is said to learn from experience E

with respect to some class of tasks T and performance measure P, if its performance at tasks in T as measured by P, improves with experience E.”

• ------ Machine Learning, Tom Mitchell, 1997

learning

What is machine learning?

• the study of algorithms that

improve their performance P at some task T with experience E

• to have a well-defined learning task, we must specify:

< P, T, E >

ML example: image classification

indoor

• utdoor

ML example: image classification

• T : given new images, classify as indoor vs. outdoor
• P : minimize misclassification costs
• E : given images with indoor/outdoor labels

ML example: spam filtering

ML example: spam filtering

• T : given new mail message, classify as spam vs. other
• P : minimize misclassification costs
• E : previously classified (filed) messages

ML example: predictive text input

ML example: predictive text input

• T : given (partially) typed word, predict the word the user

intended to type

• P : minimize misclassifications
• E : words previously typed by the user

(+ lexicon of common words + knowledge of keyboard layout) domain knowledge

ML example: Netflix Prize

• T : given a user/movie pair, predict the user’s rating (1-5

stars) of the movie

• P : minimize difference between predicted and actual

rating

• E : histories of previously rated movies

(user/movie/rating triples)

ML example: Netflix Prize

ML example: autonomous helicopter

video of Stanford University autonomous helicopter from http://heli.stanford.edu/

ML example: autonomous helicopter

• T : given a measurement of the helicopter’s current

state (orientation sensor, GPS, cameras), select an adjustment of the controls

• P : maximize reward (intended trajectory + penalty

function)

• E : state, action and reward triples from previous

demonstration flights

Google DeepMind's Deep Q-learning playing Atari Breakout From the paper “Playing Atari with Deep Reinforcement Learning”, by Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, Daan Wierstra, Martin Riedmiller

ML example: Atari Breakout

ML example: AlphaGo

Assignments

Reading assignment

• read
• Chapter 1 of Murphy
• article by Jordan and Mitchell on course website
• course website:

http://pages.cs.wisc.edu/~yliang/cs760_spring20/

HW1: Background test

• posted on course website
• will set up how to submit the solutions on Canvas
• Two parts: minimum and medium tests
• if pass both: in good shape
• if pass minimum but not medium: can still take but

expect to fill in background

• if fail both: suggest to fill in background before taking

the course

Minimum background test

• 80 pts in total; pass: 48pts
• linear algebra: 20 pts
• probability: 20 pts
• calculus: 20 pts
• big-O notations: 20 pts

Minimum test example

Minimum test example

Medium background test

• 20 pts in total; pass: 12 pts
• algorithm: 5 pts
• probability: 5 pts
• linear algebra: 5 pts
• programming: 5 pts

Medium test example

Medium test example

THANK YOU

Some of the slides in these lectures have been adapted/borrowed from materials developed by Mark Craven, David Page, Jude Shavlik, Tom Mitchell, Nina Balcan, Elad Hazan, Tom Dietterich, and Pedro Domingos.